The document discusses research on online gambling behavior. It summarizes previous research that used daily aggregates from online gambling data to identify behavioral markers of risky gambling. It then describes current research analyzing behavioral data from PlayNow, British Columbia's online gambling platform. This research examines behavioral indicators like chasing losses by increasing bet sizes. It finds some players exhibit chasing behaviors more frequently than others. The presentation outlines challenges in identifying at-risk players and areas for further research, such as predictive modeling and analyzing speed of betting.
2. The Centre for Gambling Research at UBC is
supported by the British Columbia Lottery
Corporation and the Province of BC Government.
This project received additional support from the
British Columbia Ministry of Finance Gambling
Policy and Enforcement Branch.
Disclosure
3. Online Gambling in British Columbia
Previous research into online gambling
A trial-by-trial approach: Chasing
Challenges & Outlook
Agenda
5. The PlayNow Platform: Online gambling
platform from BCLC for BC and Manitoba.
14
Source: Online Gambling in British Columbia, Lesch, Limbrick-Oldfield, Clark, 2017
6. Behavioural measures differ between games
types for median & engaged users.
Source: Online Gambling in British Columbia, Lesch, Limbrick-Oldfield, Clark, 2017
7. The majority of players access PlayNow to
place lottery bets.
3
Top 20% most engagement players
10. Data Set
– Daily aggregates from bwin based in Austria, website world wide
available (2005-2007)
– Games Types: sports betting (most), internet poker, casino
Measures
– Daily aggregates of
• number of bets, bets per day, Euros per bet, total wagered, net loss,
percent lost
– Number of active days before the first deposit was made
– Duration of play, days active
– Monetary deposits to, and withdrawals from player’s account
– Trajectory of first month wagers
– Reason for account closure
7
Previous research on online gambling relied
on daily aggregates & quantitative measures.
11. Behavioural clustering of account closures.
Source: How do gamblers start gambling: identifying
behavioural markers for high-risk internet gambling
(Braverman & Shaffer, 2010
11
530 players
closed their
account
No interest in
gambling
Due to gambling
related problems
Not satisfied with
service
378
(71%)
15 (3%)
22 (4%)
115
(22%)
Clustering on first month behaviour
Moderate Betting
High Intensity,
low variability
Low first month activity
High intensity & variability
High accordance for
closing due to gambling
related problems
33%
48%
19%
Reason for account closure
12. Decision Tree classification of RG-grouped vs.
non-grouped customers.
Source: Using Cross-Game Behavioral
Markers for Early Identification of High-Risk
Internet Gamblers (Braverman et al. 2013,
similar: Gray et al. 2012)
12
All
> 138
> 45.5
Number of
gambling
activities <2,
2 or >2
Live action
staked
variability
>138
<138
Casino
stakes
variability
<45.5
>45.5
50/50
3037
% N
Control 10 17
Target 90 158
Total 100 175
% N
Control 9 12
Target 91 116
Total 100 128
2
>2
Not high
risk
Not high
risk
Not high
risk
HIGH
RISK
HIGH
RISK
14. Behavioural markers of online gambling:
e.g. Chasing
’Loss chasing’: Trying to ‘win’ back previously lost funds.
Increased bet sizes or prolonged betting after a series of losses in
an attempt to win back funds (Lesieur, 1984; American Psychiatric
Association, 2013).
Operationalisations:
- increase bet size
- accelerate betting
- play longer
- play quicker again
15. Over the course of a gambling session,
people increase their wager about 40%.
Relative time in session
Wagerrelativetosessionstart
Average amount bet throughout session
17. On slots machines, on average there is no
bias.
Frequency
Increase of wagerDecrease of wager
Zero Line
(No Bias)
Median (No
Bias)
Distribution of Correlation Coefficients
Correlation coefficient (chasing)
18. Session outcome
Correlation
>.6
.6 < >.2
0 < >.2
0 > < -.2
-.2 > < -.6
< -.6
There is no simple relationship between chasing
and accumulated wager or session outcome.
19. Chasing (wager increase) shows limited
relationship to other measures of gambling.
all slots tables
Number of bets .019 .074 -.065
Accumulated wager .102 .122 .050
Session outcome .105 .108 -.007
Correlation of Chasing with aggregated measures of gambling
session wager session Outcomenumber of bets
wager size correlation
All Slots Mixed Tables
20. Number of chasing sessions (>.5) by user
NumberofUsers
Distribution of chasing session (>.5) by user
Some users show larger numbers of chasing
sessions.
21. Percentage of chasing sessions (>.5) by user
NumberofUsersA subset of users appear to show chasing on
almost every gambling session.
Distribution of chasing session (>.5) in % by user
22. Learning & next steps
Learnings:
• No one size fits all
– Different games require different measures (e.g. bet size
variance slots vs. tables)
– Varying consistency within and between people
• Differentiate average and extreme effects
• Limited relation of previous aggregate measures
Next steps:
• Winning vs. losing
• Look at subsets of players
• Operationalisation of chasing:
• Within vs. between players
- accelerate betting
- play longer
- play quicker again
24. Identifying at risk players requires knowing
who at risk players are.
- ’Let the data speak for itself’ (clustering – unsupervised
learning)
- What are the ‘right’ measures/markers for problematic gambling?
- What do any method’s results have to say about real behaviour?
- ’Train the data to identify certain individuals’ (classification –
supervised learning)
- Who are individuals with problematic behaviours?
- Samples of account closures, voluntary self-excluders, etc. can
provide some external confirmation.
25. At the moment, it’s all about the measures.
How can we identify problematic play.
- Additional behavioural measures
- Speed of play
- Streak- & sequence effects
- Predictive modelling
- Log-On and Log-Off
- Choice of game, game switches
- Wager size
- The People Dilemma
- Getting the getting people with the right skills.
26. Acknowledgements
CGR
Professor Luke Clark
Dr Eve Limbrick-Oldfield
www.cgr.psych.ubc.ca
@CGR_UBC
BCLC
Dr Kahlil Philander
Bradley Bodenhamer
Michaela Becker
Questions?
Dr Tilman Lesch
Tilman.Lesch@psych.ubc.ca
Thank you for your attention!
27.
28. Areas of online gambling research
1. Descriptive analysis of online gambling
– Prevalence study
– Player segmentation
– Comparative analysis within individuals, e.g. daily, monthly,
seasonal, yearly playing patterns
2. Identification of at risk players (e.g. Harvard’s Transparency Project)
– Personalized interventions
– Timely interventions
– Predictive Analytics
• Determine likelihood of problem gambling event, e.g money
upload, self exclusion, loss chasing
4
Confirmation of theory and laboratory findings in naturalistic data
30. Motivation to study online gambling
• Rapid growth since early 2000s
• Ubiquitous 24/7 availability
• Different types of players
• Much easier and quicker feedback to changes in
regulation
• Easier access to playing data
• Easy Implementation of additional measures such as
questionnaires, etc. possible
4
31. Cognitive Biases in Gambling
“ A cognitive bias is a pattern of deviation in judgment and decision-making,
whereby inferences about situations and other people may be drawn in an
illogical fashion.” Hot Hand Fallacy:
fallacious belief that a person who has experienced
success with a random event has a greater chance
of further success in additional attempts.
Illusion of Control:
tendency for people
to overestimate their
ability to control
events.
Sequential/ Streak Effects:
(“Gambler’s Fallacy”: mistaken
belief that, if something happens
more (less) frequently than
normal during some period, it will
happen less (more) frequently in
the future - balancing).
Cognitive distortions play an important role in the development and
maintenance of pathological gambling.
32. Behavioural markers of online
gambling II: Betting Speed
Translation for online gambling:
Time between one bet and the next bet within the same session.
34. Mean Betting Speed difference between
winning and losing by individual
Zero Line
(No Bias)
Slower After WinsSlower after Losses
26
>0:
Slots: ~75%
Tables: ~65%